libobs: Add packed yuv to planar 4:4:4 conversion

libobs/media-io/format-conversion.c

@@ -36,6 +36,17 @@ do {                                                                          \
 	*(uint32_t*)(lum_plane+lum_pos1) = get_m128_32_1(pack_val);           \
 } while (false)
 
+#define pack_val(lum_plane, lum_pos0, lum_pos1, line1, line2, mask)           \
+do {                                                                          \
+	__m128i pack_val = _mm_packs_epi32(                                   \
+			_mm_and_si128(line1, mask),                           \
+			_mm_and_si128(line2, mask));                          \
+	pack_val = _mm_packus_epi16(pack_val, pack_val);                      \
+                                                                              \
+	*(uint32_t*)(lum_plane+lum_pos0) = get_m128_32_0(pack_val);           \
+	*(uint32_t*)(lum_plane+lum_pos1) = get_m128_32_1(pack_val);           \
+} while (false)
+
 #define pack_ch_1plane(uv_plane, chroma_pos, line1, line2, uv_mask)           \
 do {                                                                          \
 	__m128i add_val = _mm_add_epi64(                                      \
@@ -152,6 +163,45 @@ void compress_uyvx_to_nv12(
 	}
 }
 
+void convert_uyvx_to_i444(
+		const uint8_t *input, uint32_t in_linesize,
+		uint32_t start_y, uint32_t end_y,
+		uint8_t *output[], const uint32_t out_linesize[])
+{
+	uint8_t  *lum_plane   = output[0];
+	uint8_t  *u_plane     = output[1];
+	uint8_t  *v_plane     = output[2];
+	uint32_t width        = min_uint32(in_linesize, out_linesize[0]);
+	uint32_t y;
+
+	__m128i lum_mask = _mm_set1_epi32(0x0000FF00);
+	__m128i u_mask   = _mm_set1_epi32(0x000000FF);
+	__m128i v_mask   = _mm_set1_epi32(0x00FF0000);
+
+	for (y = start_y; y < end_y; y += 2) {
+		uint32_t y_pos        = y      * in_linesize;
+		uint32_t lum_y_pos    = y      * out_linesize[0];
+		uint32_t x;
+
+		for (x = 0; x < width; x += 4) {
+			const uint8_t *img = input + y_pos + x*4;
+			uint32_t lum_pos0  = lum_y_pos + x;
+			uint32_t lum_pos1  = lum_pos0 + out_linesize[0];
+
+			__m128i line1 = _mm_load_si128((const __m128i*)img);
+			__m128i line2 = _mm_load_si128(
+					(const __m128i*)(img + in_linesize));
+
+			pack_shift(lum_plane, lum_pos0, lum_pos1,
+					line1, line2, lum_mask, 1);
+			pack_val(u_plane, lum_pos0, lum_pos1,
+					line1, line2, u_mask);
+			pack_shift(v_plane, lum_pos0, lum_pos1,
+					line1, line2, v_mask, 2);
+		}
+	}
+}
+
 void decompress_420(
 		const uint8_t *const input[], const uint32_t in_linesize[],
 		uint32_t start_y, uint32_t end_y,

libobs/media-io/format-conversion.h

@@ -37,6 +37,11 @@ EXPORT void compress_uyvx_to_nv12(
 		uint32_t start_y, uint32_t end_y,
 		uint8_t *output[], const uint32_t out_linesize[]);
 
+EXPORT void convert_uyvx_to_i444(
+		const uint8_t *input, uint32_t in_linesize,
+		uint32_t start_y, uint32_t end_y,
+		uint8_t *output[], const uint32_t out_linesize[]);
+
 EXPORT void decompress_nv12(
 		const uint8_t *const input[], const uint32_t in_linesize[],
 		uint32_t start_y, uint32_t end_y,